The latest updates on your projects. Learn more about Vercel for Git ↗︎

Thank you for your contribution! ❤️

You can try out this pull request locally by installing Rollup via

npm install rollup/rollup#feat/treeshakes-unary

Notice: Ensure you have installed the latest stable Rust toolchain. If you haven't installed it yet, please see https://www.rust-lang.org/tools/install to learn how to download Rustup and install Rust.

or load it into the REPL:
https://rollup-glzkgf4l8-rollup-js.vercel.app/repl/?pr=5775

Performance report

• BUILD: 7291ms, 734 MB
  • initialize: 0ms, 27.2 MB
  • generate module graph: 2736ms, 550 MB
    • generate ast: 1240ms, 543 MB
  • sort and bind modules: 393ms, 593 MB
  • mark included statements: 4167ms, 734 MB
    • treeshaking pass 1: 1368ms, 695 MB
    • treeshaking pass 2: 666ms, 723 MB
    • treeshaking pass 3: 275ms, 723 MB
    • treeshaking pass 4: 255ms, 725 MB
    • treeshaking pass 5: 299ms, 729 MB
    • treeshaking pass 6: 243ms, 730 MB
    • treeshaking pass 7: 226ms, 731 MB
    • treeshaking pass 8: 220ms, 733 MB
    • treeshaking pass 9: 205ms, 730 MB
    • treeshaking pass 10: 206ms, 733 MB
    • treeshaking pass 11: 205ms, 734 MB
• GENERATE: 692ms, 973 MB
  • initialize render: 0ms, 869 MB
  • generate chunks: 71ms, 879 MB
    • optimize chunks: 0ms, 874 MB
  • render chunks: 606ms, 948 MB
  • transform chunks: 15ms, 973 MB
  • generate bundle: 0ms, 973 MB

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.01%. Comparing base (f5c349e) to head (97e4b83).
Report is 2 commits behind head on master.

@@           Coverage Diff           @@
##           master    #5775   +/-   ##
=======================================
  Coverage   99.01%   99.01%           
=======================================
  Files         259      259           
  Lines        8090     8114   +24     
  Branches     1363     1370    +7     
=======================================
+ Hits         8010     8034   +24     
  Misses         53       53           
  Partials       27       27           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

I've been struggling with this for a long time, and I feel that the smallest representation can be delegated to specialized compression tools like Terser. Therefore, I feel that we just need to focus on improving tree-shaking and making the resulting code as straightforward and easy to understand as possible. Because of this, I've retained true, false, undefined, and -Infinity. However, when a UnaryExpression evaluates to a number, I handle it specifically to prevent cases where scientific notation numbers are converted into very long numbers. Furthermore, I feel we don't really need to consider shortening literals in the Literal nodes. Lukas, what do you think about this? I'd love to hear your thoughts.

I've been struggling with this for a long time, and I feel that the smallest representation can be delegated to specialized compression tools like Terser. Therefore, I feel that we just need to focus on improving tree-shaking and making the resulting code as straightforward and easy to understand as possible. Because of this, I've retained true, false, undefined, and -Infinity. However, when a UnaryExpression evaluates to a number, I handle it specifically to prevent cases where scientific notation numbers are converted into very long numbers. Furthermore, I feel we don't really need to consider shortening literals in the Literal nodes. Lukas, what do you think about this? I'd love to hear your thoughts

That sounds good to me, and I see your point with numbers. And thank you for taking the time to think this through.

This PR has been released as part of rollup@4.30.0. You can test it via npm install rollup.

Hi @TrickyPi.

In updating my projects from a previous version of Rollup I noticed that this PR converts void 0 to undefined in build output. I personally out of historical / creature of habit reasons have a lot of code that uses void 0. I'm well aware that using void 0 vs undefined doesn't have the same ancient use case that formed my habit from years past. It's just a preference now preferring slightly more compact source code / readability reasons. So, I'd like to understand exactly why the void 0 -> undefined transformation is useful in the context of Rollup. IE I don't like that quite a bit of project / framework code in source form has void 0, but the build artifacts have undefined instead. It definitely is the 4.30.0 release and this PR that changed this behavior.

Thanks for the insight and your work on Rollup.

It's just a preference now preferring slightly more compact source code / readability reasons. So, I'd like to understand exactly why the void 0 -> undefined transformation is useful in the context of Rollup.

It's just a final output representation. While choosing between void 0 and undefined to represent the result of a unary expression that might be undefined (although currently, only void operations produce undefined), I ended up prioritizing readability.

That's the thing. Readability is subjective. In my larger / complex project with many build assets with some that are intermediary / checked in to repos I immediately noticed this change. I'm at the end release candidate stage with many pre-releases prior just before a final release for this effort and updated Rollup from 4.29.x to latest for the last RC and this was a concern that needed to be investigated when every build artifact had unexpected changes.

From a readability point of view for those that use void 0 instead of undefined the annoyance is that for end devs looking at the build assets in IDEs they may be slightly confused when turning to the actual source repos. If there was a functional reason for the transformation that is understandable. A transformation like this might better be handled as opt-in otherwise, but I also grant that I'm likely in a small minority on the readability preference of void 0 over undefined.

@typhonrt Changed in this PR #5838

Awesome.. Thanks for the consideration!